The Role of Pleistocene Refugia and Rivers in Shaping Gorilla Genetic Diversity in Central Africa

The role of Pleistocene refugia and rivers in shaping gorilla genetic diversity in central Africa

Nicola M. Anthony*†‡, Mireille Johnson-Bawe§, Kathryn Jeffery†¶, Stephen L. Clifford§, Kate A. Abernethy§¶, Caroline E. Tutin§¶, Sally A. Lahmʈ, Lee J. T. White**, John F. Utley*, E. Jean Wickings§, and Michael W. Bruford†

*Department of Biological Sciences, University of New Orleans, New Orleans, LA 70148; †School of Biosciences, Cardiff University, Cardiff CF10 2TL, United Kingdom; §Centre International de Recherches Me´ dicales de Franceville, B.P. 769, Franceville, Gabon; ¶School of Biological and Environmental Sciences, University of Stirling, Stirling FK9 4LA, United Kingdom; ʈInstitut de Recherche en Ecologie Tropicale, B.P. 180, Makokou, Gabon; and **Wildlife Conservation Society, Bronx Zoo, 185th Street and Southern Boulevard, Bronx, NY 10460

Edited by Jeffrey Rogers, Southwest Foundation for Biomedical Research, San Antonio, TX, and accepted by the Editorial Board November 6, 2007 (received for review May 23, 2007) The role of Pleistocene forest refugia and rivers in the evolutionary refugia in geographical speciation (29, 30). However, it could diversification of tropical biota has been the subject of consider- also be argued that the recent time frame of Pleistocene events able debate. A range-wide analysis of gorilla mitochondrial and requires a population-genetic rather than a species-level ap- nuclear variation was used to test the potential role of both refugia proach. Another criticism centers on the real difficulties of and rivers in shaping genetic diversity in current populations. pinpointing the precise location of putative refugia (4) and Results reveal strong patterns of regional differentiation that are discriminating between competing modes of diversification (5). consistent with refugial hypotheses for central Africa. Four major Molecular data can, however, be used to infer signatures of past mitochondrial haplogroups are evident with the greatest diver- history and, in so doing, test several important predictions of gence between eastern (A, B) and western (C, D) gorillas. Coales- Pleistocene hypotheses (3, 7, 31): (i) allopatric fragmentation is cent simulations reject a model of recent east–west separation evident and coincident with the possible location of hypothesized during the last glacial maximum but are consistent with a diver- refugia; (ii) signature(s) of demographic expansion within refu- gence time within the Pleistocene. Microsatellite data also support gial populations are apparent; and (iii) haplotype exchange is a similar regional pattern of population genetic structure. Signa- coincident with boundaries of adjacent refugia. tures of demographic expansion were detected in eastern lowland In contrast, the riverine barrier hypothesis argues that tropical (B) and Gabon/Congo (D3) mitochondrial haplogroups and are rivers limit species distributions (32, 33) and shape intraspecific consistent with a history of postglacial expansion from formerly patterns of diversification (34, 35). To date, support for this isolated refugia. Although most mitochondrial haplogroups are hypothesis is equivocal and depends heavily on the ecology and regionally defined, limited admixture is evident between neigh- dispersal abilities of individual taxa. In central Africa, the Sanaga boring haplogroups. Mantel tests reveal a significant isolation-by- River is an important biogeographic boundary for several distance effect among western lowland gorilla populations. How- groups, including some primates and forest duikers (25, 36). For ever, mitochondrial genetic distances also correlate with the chimpanzee subspecies, however, this proposed barrier may be distance required to circumnavigate intervening rivers, indicating incomplete (37). Other studies have also reported that rivers a possible role for rivers in partitioning gorilla genetic diversity. influence genetic differentiation between mandrill (21) and Comparative data are needed to evaluate the importance of both mechanisms of vicariance in other African rainforest taxa. bonobo (38) populations. Where sampling is of sufficient inten- sity, molecular data can be used to test predictions of the riverine control region ͉ mitochondrial ͉ phylogeography ͉ refugium barrier hypothesis, namely, that genetic variation is structured across different river banks, and that genetic differentiation decreases from the mouth to the headwaters of a given echanisms underlying evolutionary diversification in trop- watercourse. Mical forests have intrigued biologists for more than a Gorillas are good candidates for testing geographically explicit century. Numerous hypotheses have been proposed (1–2), of hypotheses of vicariance because of their strong association with which the Pleistocene forest refugium and riverine barrier closed canopy forest and their restricted ability to traverse hypotheses have provoked considerable interest and controversy savannah–forest mosaic habitats (39). Gorillas may also be good (3–7). Whereas most studies have focused on the Amazon and models for testing the isolating effects of rivers because the Australian wet tropics (5, 8–10), data on central African rain- taxonomic boundaries of many primates appear coincident with forest taxa remain relatively sparse. According to Pleistocene refuge theory, forest fragmentation major river courses (33, 36). By drawing on data from a during glacial maxima led to the isolation and subsequent range-wide analysis of gorilla populations, we aim to uncover the diversification of forest-associated taxa (3). During periods of relative importance of rainforest refugia and rivers in shaping climate amelioration and population expansion, zones of sec- gorilla population structure and, in so doing, provide insight into ondary contact may have also formed between neighboring refugial populations. Although palynological and biogeograph- Author contributions: K.A.A., L.J.T.W., E.J.W., and M.W.B. designed research; N.M.A., ical data have been used to infer forest refugia (11–16), their M.J.-B., and S.L.C. performed research; K.J., K.A.A., C.E.T., S.A.L., L.J.TW., and J.F.U. con- precise location and role in Pleistocene diversification is con- tributed new reagents/analytic tools; N.M.A., J.F.U., and M.W.B. analyzed data; and N.M.A. troversial. Several molecular studies suggest that refugia may wrote the paper. have played an important role in structuring montane birds The authors declare no conflict of interest. (17–19), primates (20–22), and trees (23). However, other forest This article is a PNAS Direct Submission. J.R. is a guest editor invited by the Editorial Board. species such as chimpanzees (24–26) and elephants (27, 28) show Data deposition: The sequences reported in this paper have been deposited in the GenBank relatively weak regional genetic structure, suggesting that wide- database (accession nos. EU305296–EU305397). ranging and/or savannah-tolerant species may be poor indicators ‡To whom correspondence should be addressed. E-mail: [email protected]. of range changes in tropical forest cover. One criticism of the This article contains supporting information online at www.pnas.org/cgi/content/full/ Pleistocene refuge hypothesis argues that species divergence 0704816105/DC1. times often predate the Pleistocene, undermining the role of © 2007 by The National Academy of Sciences of the USA

20432–20436 ͉ PNAS ͉ December 18, 2007 ͉ vol. 104 ͉ no. 51 www.pnas.org͞cgi͞doi͞10.1073͞pnas.0704816105 12

9 28

Haplogroup A 18 Haplogroup B Haplogroup C1 Haplogroup C2 Haplogroup D1 Fig. 2. Geographic distribution of major mitochondrial haplogroups across Haplogroup D2 central Africa. Important rivers are also indicated with an arrow. CR, Cross Haplogroup D3 5 River; SA, Sanaga River; OG, Ogooue´ River; IV, Ivindo River; SG, Sangha River; 6 UB, Ubangui River; CG, Congo River. The locations of two published museum samples (22, 57) are indicated by ‘‘?.’’ 7 are in Lobe´ke´, southeastern Cameroon (site 4), and the Monts de Cristal in northwestern Gabon (site 6). Molecular diversity is greatest in western gorilla haplogroups Fig. 1. Minimum spanning tree of mitochondrial HV1 haplotypes identiﬁed C1/2, D1, and the heavily admixed Ivindo population (SI Table in all three major gorilla subspecies. The mutational steps separating haplotypes are indicated by using cross bars or by number. Major mitochondrial 1). When gorilla populations were pooled into eastern or western haplogroups (A–D) and their respective subdivisions (C1–2, D1–3) are based on groups, molecular diversity indices ␪ (40) and ␲ (41) were almost haplogroups recovered in the phylogeny (see SI Figs. 4 and 5). 2-fold higher for western gorillas (␪ ϭ 0.047, ␲ ϭ 0.058) than for eastern gorillas (␪ ϭ 0.029, ␲ ϭ 0.038). Fu’s F and Tajima’s D statistic are significantly negative for the eastern lowland gorilla historical processes that underlie the evolution of tropical forest haplogroup B and western gorilla haplogroup D3. Only Tajima’s biota. D was significant for the geographically widespread haplogroup C. Mismatch distribution profiles provide a strong unimodal Results pattern for haplogroups A, B, and D3 (data not shown). In The minimum spanning network (Fig. 1) illustrates the deep keeping with earlier recommendations (7) and the acknowl- divergence between eastern and western gorillas and the high edged upward bias in estimates of the growth parameter g (42), levels of substructuring present in western gorillas relative to only haplogroups B and D3 demonstrated significant positive EVOLUTION other gorilla subspecies. As previously observed (22), the diver- values of g that were at least three times their standard deviation. gence between eastern and western gorilla populations is far These estimates were robust to changes in initial parameter greater than that observed between mountain (A) and eastern estimates and the number and length of the Markov chains. lowland (B) gorillas. Phylogenetic analyses provide substantial Mantel tests for the association between molecular and Eu- statistical support for two regionally defined haplogroups (C, D) clidian geographic distances indicate a highly significant isola- within western gorillas [supporting information (SI) Figs. 4 and tion by distance effect within western lowland gorillas (Rey- Յ 2 ϭ ⌽ Յ 2 ϭ 5]. The first (C) extends from the Cross River area in Nigeria to nolds’s genetic distance P 0.001; r 0.496; ST P 0.001; r southeastern Cameroon and comprises two subgroups (C1, C2). 0.332). The association between Reynolds’s distance and both Whereas C1 is distributed across the entire region, C2 is limited cost matrices was also highly significant. This association was to Dja in central Cameroon, Minke´be´ in Northern Gabon, and strongest when stream orders of 6 or greater were considered Յ 2 ϭ the west bank of the Ivindo/Ayala River in central Gabon (Fig. impenetrable (P 0.001; r 0.574). In partial Mantel tests, the 2). The second major haplogroup (D) extends from coastal partial correlation between this cost matrix and Reynolds’s Gabon eastwards to Congo and the southern tip of the Central genetic distance was still significant even when controlling for the effects of Euclidian distances between sites (P Յ 0.001; r2 ϭ African Republic (CAR). Within haplogroup D, three geograph- 0.2337). MESQUITE simulations rejected the hypothesis of ically defined subgroups are evident: (i) D1 in the montane east–west divergence during the last glacial maximum but are regions of Equatorial Guinea and the adjacent Monts de Cristal consistent with a model of Pleistocene separation of 52,500– in northwestern Gabon; (ii) D2 in the Dzanga-Sangha region of 120,000 years or older. MDIV analyses support an older diver- CAR; and (iii) D3 across much of Gabon and east to Lossi, gence time of Ϸ1.6 million years, although this date still falls well Congo (Fig. 2). Although there is strong support for D2 and D3, within the Pleistocene. subgroup D1 is only weakly differentiated. Cumulative probabilities of identity for the seven nuclear Spatial analysis of molecular variation indicates that the microsatellite loci used in the present study were Ͻ0.05 in all among-group variance asymptotes at four groups (correspond- populations, even assuming full-sibling relationships. Multilocus ing to the major haplogroups A–D). With seven groups, the genotypes with less than three loci or redundant genotypes were among-group variance component reaches a near-maximal value removed from the dataset, leaving a total of 91 genotype profiles of 85.44%. These seven groupings correspond well with the with three or more typed loci (SI Table 2). Tests of Hardy– major branches recovered in phylogenetic analysis with the Weinberg equilibrium revealed no deviations after Bonferroni exception of the separation of haplogroups C1/2 and Ivindo (site correction. All pair-wise FST comparisons showed significant 19). This latter population is made up of a mixture of highly population differentiation with the exception of the two popu- divergent haplogroups (C2, D2, D3) and thus warrants separa- lations within haplogroup D3 (Lope´ and Lossi). Multidimen- tion from other population groupings. The only other areas sional scaling of microsatellite genetic distances between pop- where limited exchange between adjacent haplogroups is evident ulations from the four major haplogroups (A–D) demonstrated

Anthony et al. PNAS ͉ December 18, 2007 ͉ vol. 104 ͉ no. 51 ͉ 20433 gence of eastern and western gorillas within an early- to mid- Pleistocene time frame (e.g., refs. 46 and 47). In contrast, coalescent simulations in the present study cannot reject the hypothesis of a more recent minimum divergence time in keeping with other recent estimates based on genome-wide patterns of variation (49). It therefore seems plausible that western gorilla haplogroups arose recently during subsequent arid phases of the Pleistocene. Although much attention has focused on forest refugia in the Neotropics, surprisingly little consideration has been given to this hypothesis in equatorial Africa (43). In Africa, the lower temper- atures and greater aridity periodically experienced during the Plio-Pleistocene (50, 51) provide a historical precedent for ice age forest refugia (15). Marine sedimentary cores indicate that Ϸ2.8 Fig. 3. Multidimensional scaling plot of microsatellite distances between million years ago, the climate began to change as ice sheets became populations from all major gorilla haplogroups (A–D). In each case, the large enough to influence climate at tropical latitudes (52). This mitochondrial haplogroup afﬁliation is indicated. shift toward a cooler, drier climate has been implicated in the evolution of African bovids (53) and hominids (54). Recent geo- morphological data also provide support for lowland forest refugia strong regional differentiation at the nuclear level. In accor- during the arid phases of the Pleistocene (55). dance with the mitochondrial data, three populations from Findings from this study also suggest that major river courses haplogroup D (Lope´, Lossi, and Bai Hokou) formed the closest have played an important role in shaping boundaries of several association (Fig. 3). regional haplogroups in gorillas, notably the partial genetic Discussion structuring evident across the Sangha River (C1 and D2), the Ogooue´ River (D1 and D3), and the Sanaga River (C2 but not Our results suggest a role for both Pleistocene refugia and rivers C1). The Ivindo/Ayina River may have also influenced postgla- in structuring gorilla genetic diversity. Evidence of a Pleistocene cial expansion by directing the southern extension of haplogroup history of fragmentation and subsequent expansion is evident in C2 into northeastern Gabon. However, quantifying the degree to eastern lowland gorillas, in keeping with earlier observations which a river constitutes a barrier to gene flow is difficult without (20). However, patterns of demographic expansion were not systematically sampling from the headwaters to the estuary (e.g., observed in mountain gorillas, possibly because of the smaller refs. 8 and 35). sample sizes used here. Admixture is also evident along the Although our data support a model of regional genetic differ- borders between refugial populations, as predicted by refuge entiation consistent with the inferred location of Pleistocene refugia theory (3, 31). Whereas molecular genetic diversity is high within and distribution of major rivers, several important caveats apply. the western gorilla haplogroup C, there is very little evidence for The paucity of palynological data makes it difficult to precisely signatures of historical fragmentation and population expansion. pinpoint the location of hypothesized refugia. Rivers are also In contrast, regional allopatric differentiation within haplogroup incomplete barriers to dispersal because of seasonal variation in D is well defined. Multiple lines of evidence also indicate strong water levels and historical shifts in drainage patterns. Several signatures of population expansion in the geographically wide- alternative diversification hypotheses have been proposed (2, 6), of spread haplogroup D3. In contrast, haplogroups D1 and D2 lack which ecological gradients (EGH) (4) and riparian refugia (RRH) signatures of population expansion and appear to have a limited (16, 44, 56) have attracted considerable attention. Proponents of the geographic distribution, although this may be a limitation of the EGH argue that areas of biological diversification may coincide present sampling strategy. Nuclear microsatellite data generally with zones of ecological transition such as the savannah–forest support patterns of regional genetic differentiation observed ecotone (57) or regions of topographical complexity (58). Although between major mitochondrial haplogroups. EGH theory seems plausible for species that tolerate a broad range Although it is difficult to pinpoint the precise location of of habitats, gorillas are closed-canopy specialists whose distribution hypothesized refugia, Maley (15) provides a structural frame- does not extend across the savannah–forest boundary. Further- work in which to infer the Pleistocene history of western lowland more, patterns of genetic differentiation and/or admixture do not gorillas. The refugial origin of haplogroup D3 is most likely appear coincident with any obvious zones of ecological transition, located in the Massif du Chaillu and Mont Doudou upland although montane areas may have been important centers of refugia of southern Gabon. Similarly, haplogroup D1 is tied very diversification. In contrast, the RRH posits that tropical lowland closely to a putative upland refugium in the Monts de Cristal in taxa may have persisted during the arid phases of the Pleistocene northwestern Gabon and adjacent Equatorial Guinea. Both in riparian forest. The RRH is not new (16, 56) but remains largely upland areas are believed to have harbored species during glacial untested in African ecosystems. According to this hypothesis, maxima and are centers of diversity for several endemic plant genetic variation should be partitioned by major watersheds and not groups (13, 14, 43). Although haplogroup D2 is not associated by candidate refugia. Aside from the Dzanga-Sangha region in with any upland region, its distribution may have nevertheless CAR, the present data do not support this hypothesis, because coincided with a fluvial refugium during the arid phases of the rivers appear to limit the distribution of genetic diversity rather than Pleistocene (22, 44). The widespread distribution of a relatively act as centers of evolutionary diversification. Geo-referenced sam- small number of haplotypes within haplogroup D3 suggests a ples of other rainforest mammals are needed in order to gain a more history of past bottleneck(s) and rapid postglacial expansion. complete understanding of the most important mechanisms of This observation is reinforced by strong signatures of population diversification in central African rainforests. expansion. Although it has been suggested that forest refugia may have Materials and Methods been reservoirs of ancestral diversity rather than engines of This study features previous data (22) and uses a larger sample collected from diversification (45), we find little evidence to support the former 29 sites across the range of all three recognized subspecies: western lowland hypothesis. Although it is difficult to date haplogroup divergence Gorilla gorilla gorilla, eastern lowland G. g. graueri, and mountain G. g. times with certainty, most previous estimates place the diver- beringei gorillas (SI Table 3). Fecal and hair collections were made from sites

20434 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0704816105 Anthony et al. across the core range of the western lowland gorilla, including putative correspond to divergence times dating from the Wisconsin glacial Ϸ20,000 refugia in Gabon (sites 6, 11, and 14) and a new western gorilla population in years ago to a mid-Pleistocene interglacial period of 610,000 years, respec- Cameroon (site 2) (59). Generally only one sample per nest site was sequenced, tively (77). For all simulations, the branch length of the root was set at 5,000 whereas multiple nests per site were genotyped at microsatellite loci. DNA generations. Estimates of ␪ were derived for all data and for eastern and extraction, PCR protocols, and primers were carried out as previously de- western populations separately by using Watterson’s method (40). Upper scribed (22). Approximately 260 bp of the first hypervariable domain (HV1) of (30,000) and lower (13,400) estimates of ancestral Ne were obtained from the the mitochondrial control region were amplified from 185 gorilla samples. relationship ␪ ϭ 2Ne␮ by using control region mutation estimates of either PCR products from 55 of these samples were cloned into the TA cloning vector 0.075 ϫ 10Ϫ6 or 0.165 ϫ 10Ϫ6 substitutions per site per year, respectively (78), (Invitrogen) before sequencing. All data from this study were then combined and a generation time of 15 years. For each divergence scenario, 100 gene with sequence data published elsewhere (20, 22, 48, 60–62). matrices were simulated by using a model constrained to fit the two-refuge The complete database of sequences (n ϭ 249) was aligned in Clustal X (63), hypothesis. Sequences were simulated by using a general time-reversible and a hypervariable polycytosine stretch of the HV1 domain was deleted from model of nucleotide substitution selected by ModelTest with rates (1.0, the alignment. Model Test v.3.06 (64) was used to estimate the nucleotide 14.5442, 0.0, 0.0, 14.5442, 1.0), a proportion of invariant characters set to 0.0, substitution model that best approximated the data. By using previously and a gamma distribution with shape parameter of 0.3655 and four rate described methods (65), candidate nuclear translocations (n ϭ 51), in vitro PCR categories. From these simulated gene matrices, consensus trees were esti- recombinants (n ϭ 6), and cloned singletons (48) or variants differing by 2–5 mated by using the maximum likelihood method implemented in PAUP under bp outside the polyC (n ϭ 6) were removed. GenBank sequences without the same model parameters, except that 0.0 rate category was adjusted to geographic information were also removed (n ϭ 20), leaving 166 gorilla HV1 0.0001 because of program constraints. Slatkin and Maddison’s S value (79) sequences for analysis. Phylogenetic analysis was carried out by using PAUP 4.0 was then estimated from each consensus tree to generate a null distribution (66). Maximum likelihood estimation of phylogeny was carried out on unique of these values. This distribution was then compared with the value for the mitochondrial haplotypes (n ϭ 60). Bootstrap consensus trees were con- observed genealogy (S ϭ 1), as is the case for reciprocal monophyly between structed with 500 replicates. eastern and western populations. The two-refuge model for a given diver- A spatial analysis of molecular variance (SAMOVA) was used to estimate the gence times was accepted if S of 1 values from the coalescent simulations were number of geographically cohesive groups of populations that were maxi- observed in at least five or more cases (P Ն 0.05). East–west divergence times mally differentiated from one another (67). Only populations with three or were also estimated from MDIV (80) by using five Markov chains of 2,000,000 more samples (n ϭ 20) were considered for spatial and demographic analyses. cycles and a burn in time of 500,000 cycles. The population divergence time By using SAMOVA, populations were partitioned into regionally defined (t) was estimated from the highest posterior probability value of the parameter haplogroups (n ϭ 7), and molecular diversity estimates were obtained by using T, an effective population size of 30,000, and a generation time of 15 years. ARLEQUIN v.3.0 (68). ARLEQUIN was also used to assess signatures of demo- Microsatellite multilocus genotypes based on seven human loci (D1S550, graphic expansion by using mismatch distribution profiles (69), Tajima’s D (70), D4S1627, D5S1457, D7S817, D8S1179, D21S11, FGA) and a sex-specific and Fu’s F (71) statistic. Maximum likelihood estimates of the growth param- amelogenin locus (81) were amplified in two separate quadriplexes by using eter (g) and ␪ within regional haplogroups was estimated by using FLUCTUATE hair samples collected from site 1 (n ϭ 16), site 12 (n ϭ 72), site 23 (n ϭ 40), site (42). An initial ␪ value of 1.0 was used with a random starting tree and 24 (n ϭ 16), site 25 (n ϭ 32), and site 28 (n ϭ 18). Because of the risk of allelic transition–transversion ratio of 10. Mantel tests (72) for association between dropout and false alleles, each sample was genotyped four times. Heterozy- geographic and genetic distances among western gorilla populations (n ϭ 16) gotes were accepted if both alleles were typed twice (82), whereas homozy- were performed by using the software IBD v.1.52 (73). Mitochondrial genetic gotes were only accepted if replicated three times. Although the number of replicates is less than recommended (82), allelic dropout rates estimated from distance matrices were based on either ⌽ST values (74) or Reynolds’s transfor- mation (75). The geographic distance matrix was constructed from either the these loci indicate that only three replicates are required to give a 95% or

Euclidian distance between sites or a ‘‘cost’’ value representing the distance greater probability of a homozygote being typed correctly (83). Cumulative EVOLUTION required to circumnavigate intervening riverine barriers to gene flow. This probabilities of identity across all loci assuming either Hardy–Weinberg fre- cost matrix was also used as an indicator matrix in partial Mantel tests of the quencies or full-sibling relationships were calculated for each population by association between genetic and Euclidian distance. using GENECAP (84). To exclude the risk of sampling the same individual more Geospatial processing and analyses were carried out by using ArcEditor and than once, redundant genotypes were removed where there was an identical the Spatial Analyst extension of ArcGIS 9.0. To construct the cost matrix, digital match between multilocus genotypes. Tests for deviations from Hardy– elevation data (DEMs), processed to remove No Data voids, were obtained Weinberg, pair-wise FST calculations, and population differentiation were ␦␮ 2 from the Consultative Group on International Agricultural Research Consor- conducted in ARLEQUIN. Goldstein’s ( ) (85) and Cavalli-Sforza and Ed- tium for Spatial Information Archives. DEMs were merged into a single raster, wards’ chord (86) distances between populations were calculated by using the clipped to the boundary of the study area, and then resampled to a resolution program MSA3.0 (87). Multidimensional scaling of this distance matrix was of 0.002°. Sinks in the resulting raster were filled, and a flow accumulation carried out by using the NCSS v.6 statistical package (NCSS). raster was produced. Cells with a flow accumulation value Ն100 were ex- tracted and converted to a stream network raster. The stream raster was used ACKNOWLEDGMENTS. We thank all of the collectors who provided samples to calculate stream order (1–8) which was then used to produce a cost raster, for this and earlier work (22): M. Bermejo (University of Barcelona, Barcelona, where streams of orders 6–8 were designated as impenetrable barriers (SI Fig. Spain), M. Charpentier (Centre International de Recherches Me´ dicales de 6) and those of orders 5, 4, and 3 were assigned a cost of 50, 40, and 30, Franceville), E. Dimoto (Centre International de Recherches Me´ dicales de respectively. Streams of order 1 or 2 and areas with a flow accumulation of Franceville), J. T. Dinkagadissi (Centre International de Recherches Me´ dicales de Franceville), M. Goldsmith (Tufts University, Boston, MA), J. Groves (Wild- Ͻ100 were assigned a cost of 1. An alternative cost matrix was also con- life Conservation Society, Limbe, Cameroon), D. Idiata (Direction de la Favre et structed, where only stream orders 7 and 8 were considered impenetrable, de la Chasse, Libreville, Gabon), S. Latour (Wildlife Conservation Society, and stream order 6 was assigned a cost of 60. Libreville, Gabon), F. Maisels (Wildlife Conservation Society, Libreville, Ga- The coalescent simulation package in MESQUITE 1.05 (76) was used to bon), K. McFarland (City University of New York, New York), Y. Mihindou compare hypotheses of minimum divergence times between eastern and (Wildlife Conservation Society, Monitoring of Illegal Killing of Elephants, western gorillas. The steps described here were modified from previous work Libreville, Gabon), E. Nwufoh (Cross River National Park, Nigeria), J. Oates (City (77) and aim to estimate the minimum divergence time required to achieve University of New York), J. Okouyi (Institut de Recherche en Ecologie Tropi- cale), I. Omari (Institute Congolais pour la Conservation de la Nature, Kinshasa, reciprocal monophyly between eastern and western gorillas by using a model Democratic Republic of Congo), R. Parnell (Wildlife Conservation Society, of nucleotide substitution and an effective population size (Ne) estimated Libreville, Gabon), P. Peignot (Centre International de Recherches Me´ dicales from the data. A simple two-refuge model was constructed, where branch de Franceville), M. E. Rogers (University of Edinburgh, Edinburgh, U.K.), and E. lengths were varied from between 1,333 and 40,666 generations. Assuming Williamson (University of Stirling). This work was funded by Darwin Initiative an average generation time for gorillas of 15 years (49), these branch lengths Grant 08/044 and the University of New Orleans.

1. Bush MB (1994) J Biogeogr 21:5–17. 6. Moritz C, Patton JL, Schneider CJ, Smith TB (2000) Annu Rev Ecol Syst 31:533–563. 2. Haffer J (1997) Biodiv Conserv 6:451–476. 7. Lessa EP, Cook JA, Patton JL (2003) Proc Natl Acad Sci USA 100:10331–10334. 3. Haffer J (1969) Science 165:131–137. 8. da Silva MNF, Patton JL (1993) Mol Phylogenet Evol 2:243–255. 4. Endler JA (1982) in Biological Diversiﬁcation in the Tropics, ed Prance GT, (Columbia 9. Schneider CJ, Smith TB, Larison B, Moritz C (1999) Proc Natl Acad Sci USA 96:13869– Univ Press, New York), pp 641–657. 13873. 5. Patton JL, da Silva MNF (1998) in Endless Forms: Species and Speciation, eds Howard 10. Hugall A, Moritz C, Mousalli A, Stanisic J (2002) Proc Natl Acad Sci USA 99:6112–6117. DJ, Berlocher SH (Oxford Univ Press, New York), pp 202–216. 11. Diamond AW, Hamilton AC (1980) J Zool 191:379–402.

Anthony et al. PNAS ͉ December 18, 2007 ͉ vol. 104 ͉ no. 51 ͉ 20435 12. Mayr E, O’Hara RJ (1986) Evolution (Lawrence, Kans) 40:55–67. 49. Thalmann O, Fischer A, Lankester F, Paä¨ bo S, Vigilant L (2007) Mol Biol Evol 24:146–158. 13. Sosef MSM (1994) Wageningen Agricultural University Papers, 94-1: Studies in Bego- 50. Hamilton AC, Taylor D (1991) Clim Change 19:65–78. niaceae V. (Wageningen Agricultural University, Wageningen, The Netherlands). 51. Bonnefille R, Roeland JC, Guiot J (1990) Nature 346:347–349. 14. Rietkirk M, Ketner P, De Wilde, JJFE (1996) in The Biodiversity of African Plants,edvan 52. deMenocal PB (1995) Science 270:53–58. der Maesen LJG, (Kluwer, Dordrecht, The Netherlands), pp 608–623. 53. Vrba ES (1995) in Palaeoclimate and Evolution, with Emphasis on Human Origins, eds 15. Maley J (1996) Proc R Soc Edinburgh 104B:31–73. Vrba ES, Denton GH, Partridge TC, Burckle LH (Yale Univ Press, New Haven, CT), pp 16. Leal ME (2004) PhD thesis (Wageningen University, Wageningen, The Netherlands). 385–424. 17. Roy M (1997) Proc R Soc London Ser B 264:1337–1344. 54. deMenocal PB, Bloemendal J (1995) in Paleoclimate and Evolution, with Emphasis on 18. Smith TB, Holder K, Girman D, O’Keefe K, Larison B, Chan Y (2000) Mol Ecol 9:1505– Human Origins, eds Vrba ES, Denton GH, Partridge TC, Burckle LH (Yale Univ Press, New 1516. Haven, CT), pp 262–288. 19. Bowie RCK, Fjeldsa J, Hackett SJ, Bates JM, Crowe TM (2006) Mol Phylogenet Evol 55. Nichol JE (1999) Geogr J 165:79–89. 38:171–188. 56. Aide TM, Rivera E (1998) J Biogeogr 25:695–705. 20. Jensen-Seaman MI, Kidd KK (2001) Mol Ecol 10:2241–2247. 57. Smith TB, Wayne RK, Girman DJ, Bruford MW (1997) Science 276:1855–1857. 21. Telfer PT, Souquiere S, Clifford SL, Abernethy KA, Bruford MW, Disotell TR, Sterner KN, 58. Fjeldsa J, Lovett JC (1997) Biodiv Conserv 6:325–346. Roques P, Marx PA, Wickings EJ (2003) Mol Ecol 12:2019–2024. 59. Morgan BJ, Wild C, Ekobo A (2003) Int J Primatol 24:1129–1137. 22. Clifford SL, Anthony NM, Bawe-Johnson M, Abernethy KA, Tutin CEG, White LTJ, 60. Garner KJ, Ryder OA (1996) Mol Phylogenet Evol 6:39–48. Bermejo M, Goldsmith ML, McFarland K, Jeffery K, et al. (2004) Mol Ecol 13:1551–1567. 61. Hofreiter M, Siedel H, Vigilant L (2003) Am J Phys Anthropol 121:361–368. 23. Muloko-Ntoutoume N, Petit RJ, White L, Abernethy K (2000) Mol Ecol 9:359–363. 62. Jensen-Seaman MI, Sarmiento EE, Deinard AS, Kidd KK (2004) Am J Primatol 63:139– 24. Morin PA, Moore JJ, Chakraborthy R, Jin L, Goodall J, Woodruff D S (1994) Science 265: 147. 1193–1201. 63. Thompson JD, Gibson TJ, Plewniak F, Jeanmougin F, Higgins DG (1997) Nucleic Acids 25. Gonder MK, Oates JF, Disotell TR, Forstner MRJ, Morales JC, Melnick DJ (1997) Nature Res 25:4876–4882. 388:337–338. 64. Posada D, Crandall KA (1998) Bioinformatics 14:817–818. 26. Goldberg T, Ruvolo M (1998) Mol Biol Evol 14:976–984. 65. Anthony NM, Clifford SL, Bawe-Johnson M, Abernethy KA, Bruford MW, Wickings EJ 27. Roca AL, Georgiadis N, Pecon-Slattery J, O’Brien SJ (2001) Science 293:1473–1477. (2007) Mol Phylogenet Evol 43:553–566. 28. Eggert SL, Rasner CA, Woodruff DS (2002) Proc R Soc London Ser B 269:1993–2006. 66. Swofford DL (2002) PAUP*: Phylogenetic Analysis Using Parsimony (*and Other Meth- 29. Klicka J, Zink RM (1997) Science 277:1666–1669. ods) 4.0 Beta (Sinauer, Sunderland, MA). 30. Knapp S, Mallet J (2003) Science 300:71–72. 67. Dupanloup I, Schneider S, Excoffier L (2002) Mol Ecol 11:2571–2581. 31. Hewitt GM (1996) Biol J Linn Soc 58:247–276. 68. Excoffier L, Laval G, Schneider S (2005) Evol Bioinform Online 1:47–50. 32. Wallace AR (1849) Proc Zool Soc Lond 20:107–110. 69. Rogers AR, Harpending H (1992) Mol Biol Evol 9:552–569. 33. Ayres JM, Clutton-Brock TH (1992) Am Nat 140:531–537. 70. Tajima D (1989) Genetics 123:585–595. 34. Aleixo A (2004) Evolution (Lawrence, Kans) 58:1303–1317. 71. Fu Y-X (1997) Genetics 147:915–925. 35. Peres CA, Patton JL, da Silva MNF (1996) Folia Primatol 67:113–124. 72. Mantel N (1967) Cancer Res 27:209–220. 36. Kingdon J (1995) The Kingdon Field Guide to African Mammals (Princeton Univ Press, 73. Bohonak AJ (2002) J Heredity 93:153–154. Princeton), pp 1–450. 74. Excoffier L, Smouse P, Quattro J (1992) Genetics 131:479–491. 37. Gonder MK, Disotell TR, Oates JF (2006) Int J Primatol 27:1103–1127. 75. Reynolds J, Weir BS, Cockerham CC (1983) Genetics 105:767–779. 38. Eriksson J, Hohmann G, Boesch C, Vigilant L (2004) Mol Ecol 13:3425–3435. 76. Maddison W, Maddison D (2006) Mesquite: A Modular System for Evolutionary Anal- 39. Tutin CEG, White LJT, Mackanga-Missandzou A (1997) Conserv Biol 11:1190–1203. ysis, Version 1.12 http://mesquiteproject.org. 40. Watterson GA (1975) Theor Popul Biol 7:256–276. 77. DeChaine EG, Martin AP (2006) Evolution (Lawrence, Kans) 60:1004–1013. 41. Tajima D (1993) in Mechanisms of Molecular Evolution, eds Takahata N, Clark AG 78. Pakendorf B, Stoneking M (2005) Annu Rev Genom Hum Genet 6:165–183. (Japan Sci Soc Press, Tokyo/Sinauer, Sunderland, MA), pp 37–59. 79. Slatkin M, Maddison WP (1989) Genetics 123:603–613. 42. Kuhner M, Yamato J, Felsenstein J (1998) Genetics 149:429–434. 80. Nielsen R, Wakely J (2001) Genetics 158:885–896. 43. Plana V (2004) Phil Trans R Soc London B 359:1585–1594. 81. Sullivan KM, Mannucci A, Kimpton CP, Gill P (1993) BioTechniques 15:636–641. 44. Colyn M, Gautier-Hion A, Verheyen W (1991) J Biogeogr 18:403–407. 82. Taberlet P, Griffin S, Goossens B, Questiau S, Manceau V, Escaravage N, Waits LP, 45. Evans BJ, Kelley DB, Tinsley RC, Melnick DJ, Cannatella DC (2004) Mol Phylogenet Evol Bouvet J (1996) Nucleic Acids Res 24:3189–3194. 33:197–213. 83. Jeffery K (2003) PhD thesis (Cardiff University, Cardiff, UK). 46. Ruvolo M (1996) Mol Phylogenet Evol 5:202–219. 84. Wilberg MJ, Dreher BP (2004) Mol Ecol Notes 4:783–785. 47. Jensen-Seaman MI, Deinard AS, Kidd K (2003) in Gorilla Biology: A Multidisciplinary 85. Goldstein DB, Ruiz Linares A, Cavalli-Sforza LL, Feldman MW (1995) Proc Natl Acad Sci Perspective, eds Taylor AB, Goldsmith ML (Cambridge Univ Press, Cambridge, UK), pp USA 92:6723–6727. 247–268. 86. Cavalli-Sforza LL, Edwards AWF (1967) Am J Hum Genet 19:233–257. 48. Thalmann O, Hebler J, Poinar HN, Paä¨ bo S, Vigilant L (2004) Mol Ecol 13:321–335. 87. Dieringer D, Schlotterer C (2003) Mol Ecol Notes 3:167–169.

20436 ͉ www.pnas.org͞cgi͞doi͞10.1073͞pnas.0704816105 Anthony et al.